A creative project grounded in research

Google’s October 21, 2026 account of Visionaries describes work with artist Prune Nourry and Google Arts & Culture to reconstruct historical sculptors’ studios. The accompanying documentary drew on archival research by art historians and used generative image tools where photographs of the workspaces were scarce.

The exhibition opened at the Petit Palais on October 20. We use October 21 as the event date for this article because it marks the published account of the AI workflow, rather than presenting the exhibition opening and the technical report as the same event.

Our analysis: a plausible image is not an archival record

When a visual brief contains gaps, a generative system can fill them convincingly. That is useful for illustration, but it creates an editorial responsibility: a viewer should be able to distinguish documented details from interpretation.

The same issue appears outside museums. A brand may reconstruct a discontinued product, a designer may visualize an unbuilt interior, or a publisher may illustrate a scene for which no photograph exists. In each case, keep the original reference material separate from generated versions and describe the purpose of the final image.

A practical asset record can identify which elements are supported by evidence, which were inferred, and which were intentionally invented for composition. This helps reviewers correct an attractive but inaccurate result before it enters a catalog or campaign.

For client work, agree on that boundary before generation begins. “Historically informed illustration” and “faithful restoration of a supplied photograph” imply different acceptance criteria. The distinction belongs in the brief as well as the final caption.

Archival evidence, a constrained visual brief, generated interpretation, and an approved labeled image.
XMH.NET editorial diagram: Keep documented details distinct from creative reconstruction. This is a conceptual workflow, not a provider architecture or benchmark.

A reviewable reconstruction workflow

Our suggested workflow keeps provenance visible through every revision. The goal is to make the image useful while preserving the evidence needed to explain it.

  • Collect references with source, date, permission, and known limitations.
  • List factual constraints separately from creative choices in the brief.
  • Review generated details against the references, especially objects and text.
  • Publish an appropriate caption and retain the approved version with its record.

Official sources

This article covers an AI industry event. XMH.NET specializes in image generation and editing APIs; coverage does not imply that every model, product, or feature described is available through our service.